Zammo offers in-depth text analytics, natural language processing, and machine learning capabilities out of the box to help you create rich conversational AI solutions.
In this article, we share how Zammo leverages artificial intelligence, what we do, what we don’t do, and our stance on automating learning.
Automated learning is the process of automatically performing repetitive and time-consuming tasks during the model creation phase in an AI solution.
For example, consider a Bayesian Classifier that performs sentiment analysis of social media data. To support this, a corpus with examples of positive, negative, and indifferent sentiment must initially be gathered and manually categorized.
New data can then be processed and classified by the Bayesian Classifier and sentiment can be determined. Newly classified data can (in theory), be used as an additional classification example and stored in the corpus for subsequent data processing and classification.
This is an example of automated learning and can support the implementation of a self-learning solution, without human intervention.
Automated learning has its place but reducing human involvement from the model creation or maintenance phases of your conversational AI solution can introduce unintended consequences.
For example, in 2016 Microsoft had to shut down an AI chatbot Tay. The chatbot was designed to develop conversational understanding by interacting with humans on Twitter. Users could follow and interact with Tay and it would tweet back a reply and then continually learn (automatically) via each interaction.
Bad actors on the social media platform realized they could train Tay with hateful comments and within 24 hours, the chatbot was posting hateful replies. Clearly letting AI handle everything in this example was not a good idea.